A fine-tuned variant of Qwen2.5 7B, this model carries the foundation of Alibaba's mid-sized open-weight architecture but has been adapted by avatargrim for purposes not fully documented publicly. Its 32K context window gives it reasonable room to handle longer inputs, though specifics about what the fine-tuning targeted remain unclear.